lab-ops
MCP server that converts SSH operations on training servers into AI-callable tools for GPU monitoring, job submission, file transfer, and more.
README
🧪 lab-ops
一个跑在本地的 MCP server,把训练服务器上那些反复手写的 SSH 操作封装成工具,交给支持 MCP 的 AI 助手(Claude Code 等)直接调度:查 GPU、后台提交训练任务、追踪任务状态、看日志、传文件——不用再每次现场敲命令。
服务器侧不需要装任何东西,本地通过 SSH/SCP 远程操作,密钥认证,零额外服务常驻。
✨ 特性
- 🎮 GPU 状态一览 — 显存占用、利用率,一条命令看全部卡
- 🚀 后台任务提交 — 基于
tmux,关掉终端任务也不会断 - 🧠 提交前显存预检查 — 自动解析命令里的
CUDA_VISIBLE_DEVICES,显存不够直接拒绝提交,避免又一次眼睁睁看着 OOM - 📊 任务状态追踪 —
running/completed/failed,附带真实退出码,不用再靠肉眼翻 Traceback - 📄 日志随时看 — 支持只看最后 N 行
- 📦 文件双向传输 — 上传/下载任意文件(checkpoint、推理结果图、脚本……)
- 🔒 目录白名单 — 所有远程路径操作限制在配置好的目录范围内,防止手滑
🛠️ 工具列表
| 工具 | 作用 |
|---|---|
gpu_status() |
查看各 GPU 显存占用和利用率 |
list_jobs() |
列出当前活跃的 tmux 任务 session |
submit_job(session_name, workdir, command, conda_env, min_free_mb=None, force=False) |
提交后台训练/评估任务,自带显存预检查 |
job_status(session_name, log_lines=20) |
查看任务状态 + 退出码 + 最新日志 |
kill_job(session_name) |
终止指定任务 |
tail_log(log_path, lines=100) |
查看日志最后 N 行 |
list_remote_files(remote_dir) |
列出远程目录文件(按修改时间排序) |
download_file(remote_path, local_path) |
下载远程文件到本地 |
upload_file(local_path, remote_path) |
上传本地文件到远程 |
📋 环境要求
- 本地:Python 3.10+,能通过密钥免密 SSH 到目标服务器
- 远程:
tmux、conda(或调整config.py里的激活逻辑换成你自己的环境管理方式)
🚀 快速开始
git clone <your-repo-url> lab-ops
cd lab-ops
pip install -r requirements.txt
cp config.example.json config.json
编辑 config.json,填入你自己的信息:
{
"host": "your.server.ip",
"user": "your_ssh_username",
"conda_base": "/home/your_username/anaconda3",
"ssh_timeout": 15,
"default_min_free_mb": 2048,
"remote_base_dirs": [
"/home/your_username/your_project_dir"
]
}
⚠️
config.json已被.gitignore忽略,不会被提交,放心填真实信息。
本地调试:
python server.py
# 或用 MCP inspector
mcp dev server.py
🔌 接入 Claude Code
claude mcp add lab-ops -- python /path/to/lab-ops/server.py
注册后运行 /mcp 确认 lab-ops 已连接,就可以直接用自然语言让助手帮你查 GPU、提交任务、看日志了。
🔐 安全说明
- 远程命令统一走
ssh -o BatchMode=yes,密钥认证失败会直接报错,不会卡住等密码输入 - 所有涉及路径的工具(
workdir、remote_dir、remote_path、log_path)都会校验是否在remote_base_dirs白名单内 submit_job/kill_job会先检查同名 session 是否存在,避免误覆盖或误杀正在跑的任务- 本项目不存储、不上传任何密码;SSH 认证完全基于密钥
📄 License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.